Reconfigurable Selector-Free All-Optical Controlled Neuromorphic Memristor for In-Memory Sensing and Reservoir Computing.

Recently, the rising demand for data-based applications has driven the convergence of image sensing, memory, and computing unit interfaces. While specialized electronic hardware has spurred advancements in the in-memory and in-sensor computing, integrating the entire signal-processing chain into a single device still faces significant challenges. Here, a reconfigurable all-optical controlled memristor with the selector-free feature is demonstrated. The conductance of the device can be controlled within the pure light domain, which enables it to integrate sensing, memory, and computing together. The integrate-and-fire behavior is also realized through electrical stimuli. Furthermore, the device exhibits an excellent rectifying ratio and nonlinearity to overcome the sneak current. Finally, an in-memory sensing and computing architecture is realized through reservoir computing based on neuron and synaptic functions mimicked by the proposed device. Such an all-in-one paradigm facilitates the computing architecture with low energy consumption, low latency, and reduced hardware complexity.